[test][LoongArch] Mark old JIT ExecutionEngine tests as unsupported

[xen0n]

Summary

The old MCJIT won't be accepting new architectures so it can only be
appropriate to add ourselves to the exclude list.

[SixWeining]

Do you mean that new archs will never get supported by MCJIT? Is there an announcement?

But it seems true because I find this MR for llvmpipe adding orcjit support for riscv: https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17801

[xen0n]

In D134367#3807481, @SixWeining wrote:

Do you mean that new archs will never get supported by MCJIT? Is there an announcement?

But it seems true because I find this MR for llvmpipe adding orcjit support for riscv: https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17801

I forgot where I read this, but I believe this is the case. Seeing thriving architectures like AArch64 and RISCV here also explains things...

[SixWeining]

In D134367#3814421, @xen0n wrote:

In D134367#3807481, @SixWeining wrote:

Do you mean that new archs will never get supported by MCJIT? Is there an announcement?

But it seems true because I find this MR for llvmpipe adding orcjit support for riscv: https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17801

I forgot where I read this, but I believe this is the case. Seeing thriving architectures like AArch64 and RISCV here also explains things...

I see related discussion here: https://discourse.llvm.org/t/ive-seen-orcjit-is-under-overhaul-and-also-the-mcjit-so-whats-the-plan/48995/2
But I'm not sure about this. MCJIT doesn't accept new architectures any more? What do you think? @rengolin @MaskRay @lhames

BTW, ExecutionEngine regression tests are enabled on AArch64 by a recent (last week) commit:

commit 23dda2899efc784ec8510a8a7b28901bc4c7c8b8
Author: Lang Hames <[e-mail]>
Date:   Tue Oct 4 11:31:09 2022 -0700

    [ExecutionEngine] Enable ExecutionEngine regression tests on AArch64.

diff --git a/llvm/test/ExecutionEngine/lit.local.cfg b/llvm/test/ExecutionEngine/lit.local.cfg
index 3338d19e967f..b00ef0dcbdf0 100644
--- a/llvm/test/ExecutionEngine/lit.local.cfg
+++ b/llvm/test/ExecutionEngine/lit.local.cfg
@@ -1,4 +1,4 @@
-if config.root.native_target in ['Sparc', 'PowerPC', 'AArch64', 'ARM64', 'SystemZ', 'Hexagon', 'RISCV']:
+if config.root.native_target in ['Sparc', 'PowerPC', 'SystemZ', 'Hexagon', 'RISCV']:
     config.unsupported = True

[rengolin]

As per that thread, indeed MCJIT is on the way out and you should go for OrcJIT if you want JIT support in LoongArch.

[xen0n]

rebase

[lhames]

Late to the party here, but I agree: New architectures should use JITLink (ORC's new JIT linker), rather than MCJIT.

This test directory actually covers both. The right thing to do is to disable LoongArch until you have a basic JITLink implementation written. Then you can enable it and start adding lli and llvm-jitlink tests.

[wangleiat]

Hi @lhames,

I submitted two patches for LoongArch support, could you help to review them?

D141036 (for JITLink) and D141102 (for ORC lazy).

llvm/test/ExecutionEngine/lit.local.cfg

-if config.root.native_target in ['Sparc', 'PowerPC', 'SystemZ', 'Hexagon', 'RISCV']:
+if config.root.native_target in ['Sparc', 'PowerPC', 'SystemZ', 'Hexagon', 'RISCV', 'LoongArch']:
     config.unsupported = True
 
 # ExecutionEngine tests are not expected to pass in a cross-compilation setup.
 if 'native' not in config.available_features:
     config.unsupported = True